Artificial Intelligence and Machine learning models have a large scope and application in Automotive embedded systems. These models are used in the automotive world for various applications like calibration, simulation, predictions, etc. These models are generally very accurate and play the role of a virtual sensor. However, the AI/ML models are resource intensive which makes them difficult to execute on largely optimized automotive embedded systems. The models also need to follow safety standards like ASIL-D.
The current work involves creating a Global DoE with ETAS ASCMO to generate data from a 125cc single to create AI/ML model for the engine outputs like Torque, T3, Mid-cat temperatures etc. The created models were validated across the operating space of the engine and found to have good accuracies.
With ETAS Embedded AI Coder, the torque and T3 prediction AI models were converted to embedded code which can be easily used as a virtual sensor in real time. Using these AI models, accurate predictions can be made on the ECU in real time without an actual sensor, thus paving to remove these sensors and reduce cost per vehicle.